Assess whether a generative-AI incident is a policy breach or a model defect

Situation

A business unit that already went live without a risk tier put incident log of hallucinated citations in a legal memo in front of board AI liaison in a bank preparing for a model-risk exam. This AI Governance / Vendors and Agentic Systems close is a generative-AI incident is from incident log of hallucinated citations in a legal memo, and the live options are A generative-AI incident is a policy breach, A model defect.

Decision

Board AI liaison in a bank preparing for a model-risk exam must choose A generative-AI incident is a policy breach / A model defect using incident log of hallucinated citations in a legal memo after a business unit that already went live without a risk tier.

Hypotheses to test

• A business unit that already went live without a risk tier is noise around an already-controlled Vendors and Agentic Systems process in a bank preparing for a model-risk exam, given incident log of hallucinated citations in a legal memo.
• A business unit that already went live without a risk tier is the event in incident log of hallucinated citations in a legal memo that forces A generative-AI incident is a policy breach for board AI liaison under AI Governance.
• Incident log of hallucinated citations in a legal memo shows a one-file miss after a business unit that already went live without a risk tier, not a Vendors and Agentic Systems program failure.
• Incident log of hallucinated citations in a legal memo cannot decide a generative-AI incident is yet after a business unit that already went live without a risk tier; hold is the only AI Governance close a bank preparing for a model-risk exam can defend.

Analysis required

• Split policy-or-governance failure from a model defect using prompts, outputs, and human edits in incident log of hallucinated citations in a legal memo.
• Reproduce the incident row in incident log of hallucinated citations in a legal memo and say whether it ever touched production data.
• Split policy-or-governance failure from a model defect using prompts, outputs, and human edits in incident log of hallucinated citations in a legal memo.
• For this AI Governance Vendors and Agentic Systems file, read incident log of hallucinated citations in a legal memo against a business unit that already went live without a risk tier and write the one fact that would move a generative-AI incident is for board AI liaison.